#7171st of 90 in Open modelsFor image, video and design, video and animation

magi-1

Autoregressive video generation model with open weights and inference code

Stars
3.8k
License
Apache-2.0
Last commit
Jun 2026
Language
Python
Stars, last 29 weeks+151
25 Apr 2026: 3,654 stars2 May 2026: 3,656 stars9 May 2026: 3,660 stars16 May 2026: 3,667 stars23 May 2026: 3,671 stars30 May 2026: 3,682 stars6 Jun 2026: 3,686 stars13 Jun 2026: 3,690 stars20 Jun 2026: 3,693 stars27 Jun 2026: 3,703 stars4 Jul 2026: 3,717 stars11 Jul 2026: 3,722 stars18 Jul 2026: 3,728 stars25 Jul 2026: 3,734 stars1 Aug 2026: 3,738 stars8 Aug 2026: 3,749 stars15 Aug 2026: 3,755 stars22 Aug 2026: 3,760 stars29 Aug 2026: 3,765 stars5 Sept 2026: 3,773 stars12 Sept 2026: 3,780 stars19 Sept 2026: 3,783 stars26 Sept 2026: 3,786 stars3 Oct 2026: 3,787 stars10 Oct 2026: 3,789 stars17 Oct 2026: 3,791 stars24 Oct 2026: 3,793 stars31 Oct 2026: 3,795 stars7 Nov 2026: 3,802 stars14 Nov 2026: 3,805 stars

3,654 on 25 Apr 20263,805 on 14 Nov 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

MAGI-1 generates video chunk by chunk (24 frames per chunk), with up to four chunks denoised concurrently. The repo has inference code and weights for 24B and 4.5B models, plus distilled and fp8-quantized variants. It supports text-to-video, image-to-video and video-to-video, driven by run.sh scripts and JSON configs.

Who it is for: Engineers with NVIDIA GPUs running or studying open video generation models

Strengths

  • 4.5B models run on one RTX 4090; distill+fp8 needs 12GB with window_size 1
  • Supports t2v, i2v and v2v, so existing video can be continued
  • Distilled and fp8-quantized checkpoints lower hardware needs for the 24B model
  • Apache-2.0 code, with base, distill and quant weights on Hugging Face

Weaknesses

  • 24B models need 8x H100/H800; quantized needs 4x H100 or 8x RTX 4090
  • No web UI or API server; inference runs through run.sh scripts
  • Source install pins Python 3.10, PyTorch 2.4 and CUDA 12.4
  • No tagged releases; the Docker command uses --privileged and host networking

What it needs

  • GPU required
  • Needs PyTorch 2.4.0, CUDA 12.4, ffmpeg 4.4, MagiAttention (optional, Hopper GPUs), T5 and VAE weights from Hugging Face
  • Models: MAGI-1 24B, MAGI-1 4.5B, MAGI-1.1 24B, distill, distill+fp8_quant

Also in Open models

See all 90
Also in Open models
RankProjectScore
1ruviewWiFi CSI sensing on ESP32 for presence, breathing, heart rate and pose104.5k starsMIT94 out of 100
2layaSingle-pass multilingual classifier returning typed decisions with calibrated odds Live demo ↗ (opens in a new tab)35.6k starsApache-2.088 out of 100
3rf-detrReal-time transformer models for object detection, segmentation and keypoints Live demo ↗ (opens in a new tab)10.1k starsApache-2.082 out of 100
#4timesfmPretrained time-series foundation model for zero-shot forecasting from Google Research35.6k starsApache-2.080 out of 100
#5open_clipPyTorch library for training and running CLIP-style image-text models14.3k starscustom license79 out of 100